Mcp Conductor

MCP Conductor — Multi-Server Orchestration

Overview

Teach your AI agent to orchestrate multiple MCP servers as a unified intelligence pipeline. Single MCP calls give shallow results. Chaining them gives 10x depth.

When to Use

• Task requires data from 2+ sources (search + scrape + analyze)
• User asks to "research" something (implies multi-source)
• Need to validate information across sources
• Building a comparison or competitive analysis
• Monitoring for changes over time
• Any workflow where one tool's output feeds another tool's input

Orchestration Patterns

Pattern 1: Research Pipeline

TASK: "Research how company X handles authentication"

Step 1: DISCOVER (Exa MCP)
  → web_search_exa: "company X authentication architecture blog"
  → Returns: 5-10 relevant URLs

Step 2: EXTRACT (Bright Data MCP)  
  → scrape_as_markdown: each URL from Step 1
  → Returns: full page content as markdown

Step 3: ANALYZE (Agent reasoning)
  → Read all scraped content
  → Extract: patterns, technologies, trade-offs
  → Cross-reference findings

Step 4: ENRICH (GitHub MCP / gh CLI)
  → Search for open-source implementations mentioned
  → Read relevant source code for concrete examples

Step 5: SYNTHESIZE (Agent output)
  → Combine all sources into structured analysis
  → Cite sources, compare approaches, recommend

Pattern 2: Competitive Intelligence

TASK: "Compare all tools that solve X"

Step 1: SEARCH (Exa MCP — semantic search)
  → "tools libraries frameworks for X comparison 2026"
  
Step 2: VALIDATE (GitHub — verify repos exist and are active)
  → gh repo view each candidate — check stars, last push, activity
  → Filter: only repos pushed in last 90 days with 100+ stars

Step 3: DEEP-READ (Bright Data — get full documentation)
  → Scrape README, docs pages, getting-started guides
  
Step 4: ANALYZE (Agent — structured comparison)
  → Feature matrix, trade-offs, community health, performance claims
  → Output: ranked recommendation with evidence

Pattern 3: Content Aggregation & Synthesis

TASK: "Create a comprehensive guide on topic Y"

Step 1: MULTI-SOURCE SEARCH
  → Exa: semantic search for authoritative sources
  → Bright Data search_engine: Google for official docs
  → GitHub: find repos, examples, real implementations

Step 2: PARALLEL EXTRACTION
  → Bright Data scrape_as_markdown: top 10 URLs (parallel batch)
  → gh API: read READMEs of top repos

Step 3: KNOWLEDGE GRAPH
  → Identify: key concepts, relationships, prerequisites
  → Map: which source covers which subtopic best
  → Find: gaps no single source covers

Step 4: GENERATION
  → Synthesize: original content citing all sources
  → Validate: cross-check claims against multiple sources
  → Format: structured guide with clear sections

Pattern 4: Monitor & React

TASK: "Watch for changes in domain Z and alert me"

Step 1: BASELINE (one-time)
  → Exa + Bright Data: current state of domain Z
  → Store: key facts, current leaders, latest news

Step 2: PERIODIC CHECK (via loop skill)
  → Exa: search for NEW content since last check
  → Compare: against stored baseline
  → Detect: changes, new entries, removals

Step 3: REACT (when change detected)
  → Bright Data: scrape the changed resource for details
  → Analyze: significance of change
  → Notify: user with summary + links

MCP Server Capabilities Reference

| Server | Best For | Key Tools | |--------|----------|-----------| | Exa | Semantic/neural web search | web_search_exa (conceptual matching, not just keywords) | | Bright Data | Any URL scraping, SERP, structured data | scrape_as_markdown, search_engine, web_data_* | | GitHub (gh CLI) | Repos, code, issues, PRs | gh search repos, gh api, gh repo view | | Slack | Team communication | slack_list_channels, slack_post_message | | Linear | Project management | Issues, projects, cycles | | Stripe | Payments | Customers, invoices, subscriptions |

Orchestration Rules

1. Start broad, narrow deep. First search finds candidates. Then scrape only the best.
2. Validate before extracting. Don't scrape 50 URLs — verify relevance first (check title, snippet).
3. Parallelize when possible. Multiple scrapes of independent URLs can happen simultaneously.
4. Cache aggressively. If you scraped a URL this session, don't scrape it again.
5. Cite everything. Every claim must trace back to a source URL.
6. Fail gracefully. If one MCP fails, continue with others. Don't abort the pipeline.
7. Respect rate limits. Space Bright Data calls by 1-2 seconds. Batch Exa calls.

Anti-Patterns

• Using ONLY one MCP when a chain would give 10x better results
• Scraping everything without filtering first (wastes time and tokens)
• Not cross-referencing (single source = unverified)
• Sequential when parallel is possible
• Ignoring MCP errors instead of retrying/falling back

Quick Reference: Common Chains

| Goal | Chain | |------|-------| | Research a topic | Exa search → Bright Data scrape top 5 → synthesize | | Find best tool for X | Exa search → GitHub verify → Bright Data docs → compare | | Monitor a space | Exa baseline → loop + Exa delta → Bright Data details | | Competitive analysis | Exa search → GitHub stats → Bright Data pricing pages → matrix | | Build knowledge base | Exa broad → Bright Data batch scrape → extract + organize |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Adit-Jain-srm
• Source: Adit-Jain-srm/skill-forge
• License: MIT
• Homepage: https://github.com/Adit-Jain-srm/skill-forge#quick-start

Install and usage instructions live in the source repository linked above.